Provided is a method of creating a local database for local optimization of an object detector based on a deep neural network. The method includes performing preprocessing on an image extracted from real-time collected or pre-collected images from an edge device, modeling a static background image based on the image received through the pre-processing unit and calculating a difference image between a current input image and a background model to model a dynamic foreground image, detecting an object image from the image based on a training model, and creating a local database based on the background image, the foreground image synthesized with the background image, and the object image synthesized with the background image.
Legal claims defining the scope of protection, as filed with the USPTO.
2. The system of claim 1, wherein the edge device calculates a difference image between the background model and a current input image to create the foreground image.
3. The system of claim 1, wherein the edge device creates the local database based on the background image, the foreground image synthesized with the background image, and the object image synthesized with the background image.
4. The system of claim 1, wherein the edge device collects a fixed surveillance image as the image.
6. The system of claim 5, wherein the background modeling unit receives the enhanced image or the channel-split image from the preprocessing unit and models the static background image based on n consecutive past images (n is a natural number greater than or equal to two) from a current image.
7. The system of claim 6, wherein the n past images are adjusted based on an amount of change in motion of a dynamic object.
8. The system of claim 5, wherein the detection unit arranges detected object images based on detection reliability and then transmits the object image having detection reliability greater than or equal to a threshold value to the post-processing unit.
9. The system of claim 5, wherein the post-processing unit adjusts an order of the morphology operations and the number of morphology operations based on a ratio of noise and loss of the merged image.
10. The system of claim 5, wherein the post-processing unit binarizes each pixel into a candidate group of an object and other background and noise based on a result of performing the morphology operation.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
October 8, 2021
July 2, 2024
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.